This work introduces a lightweight framework for 3D object classification in point clouds, derived from the Attention-Based Point Cloud Edge Sampling (APES) model. The approach, termed APES-Soft, is based on a systematic ablation study aimed at reducing architectural complexity while preserving classification performance. Three ablation scenarios were investigated, each altering distinct network components to determine the most effective configuration. Scenario II emerged as the most favorable trade-off variant, reaching 93.8% Accuracy alongside 93.7% Precision, 93.8% Sensitivity, 93.7% F1-Score, 93.5% Matthews Correlation Coefficient (MCC), and 89.2% Jaccard index—within the range of reported ModelNet40 results. A supplementary evaluation on ScanObjectNN PB_T50_RS further showed that Scenario II retained competitive performance on a harder benchmark, reaching 80.88%. Furthermore, Scenario II reduced training time to 20.35 hours and memory usage by 21.89%, while using 0.817M parameters, 5.885 GMACs/sample, 11.770 GFLOPs/sample, and 18.141 ms/sample for single-sample inference. Statistical analyses, including ANOVA, Tukey’s HSD, Kruskal–Wallis, and Friedman tests, were interpreted conservatively and do not support claims of formal equivalence or statistically significant superiority. Instead, the results indicate that Scenario II maintains a competitive performance profile while reducing computational cost, highlighting APES-Soft as a reliable and efficient solution for 3D object classification in resource-limited environments.

We introduce a novel technique called ShapedNet to enhance body composition assessment. This method employs a deep neural network capable of estimating Body Fat Percentage (BFP), performing individual identification, and enabling localization using a single photograph. The accuracy of ShapedNet is validated through comprehensive comparisons against the gold standard method, Dual-Energy X-ray Absorptiometry (DXA), utilizing 1273 healthy adults spanning various ages, sexes, and BFP levels. The results demonstrate that ShapedNet outperforms in 19.5% state of the art computer vision-based approaches for body fat estimation, achieving a Mean Absolute Percentage Error (MAPE) of 4.91% and Mean Absolute Error (MAE) of 1.42. The study evaluates both gender-based and Gender-neutral approaches, with the latter showcasing superior performance. The method estimates BFP with 95% confidence within an error margin of 4.01% to 5.81%. This research advances multi-task learning and body composition assessment theory through ShapedNet.

Stroke is a leading cause of death globally, with higher chances of recovery when prompt and accurate diagnosis is followed by appropriate treatment. Various neuroimaging techniques, including computed tomography (CT), are used for stroke detection. Computer-aided diagnosis (CAD) systems can capture information imperceptible to the human eye, making them valuable tools in stroke diagnosis. This study proposes a novel approach for segmenting hemorrhagic stroke in CT scans using Deep Learning. Specifically, we evaluate the effectiveness of SSD, YOLO-v4, and YOLACT as prompts for the Segment Anything Model (SAM) in hemorrhagic stroke segmentation. Additionally, we compare YOLACT and SAM for segmentation performance. The methods showed promising results, with the proposed SAM and MobileSAM achieving an accuracy of 99.82%, while YOLACT attained 99.74%. The use of zeroshot and one-stage models demonstrated exceptional efficiency in addressing the segmentation challenges in medical images.

Diabetic retinopathy (DR) is a complication resulting from diabetes mellitus, caused by damage to the blood vessels in the retina due to excess glucose in the blood. This condition is one of the leading causes of vision loss in adults with diabetes. Early detection and appropriate treatment are crucial to prevent the progression of the disease. The main objective of this study is to develop an advanced tool for classifying retinal image photographs. To achieve this, we used the Normalizer Free Neural Network (NFNet) as a feature extractor, combining it with machine learning techniques through the method of transfer learning. We evaluated the effectiveness of our model by comparing its results with those of various established and recognized convolutional neural network architectures in the literature. The results show that, using the NFNet, we achieved an accuracy of 99.83% and an F1-Score of 99.46% when combined with support vector machines using radial basis function kernel. Given the significance of these results, the next step is to explore the possibility of developing a diagnostic support tool using the developed methodology.

The advancement of machine learning techniques has brought significant progress to the classification of brain tumors, proving essential for early diagnoses and effective treatments. This study focuses on evaluating the performance of feature extractors and classifiers for the binary division of brain tumors. Five extractors were employed: Zernike Moments, DWT (Discrete Wavelet Transform), LBP (Local Binary Pattern), GLCM (Gray-Level Co-occurrence Matrix) and HU Moments. The HU Moments extractor stood out with the shortest average execution time, 13.56 microseconds. Subsequently, the Grid Search technique was employed to identify the best hyperparameters for five classifiers: Random Forest, K-Nearest Neighbors, Support Vector Machine, Multilayer Perceptron, and Naive Bayes. Four evaluation metrics were used, prioritizing precision and F1-Score. The results revealed that the combination of Zernike Moments and KNeighbors achieved a test precision of 100.00%, surpassing the results of previous studies.

Free Space Optics (FSO) has emerged as a crucial communication modality in recent years, especially with the implementation of techniques associated with the 5G network and its subsequent advancements. This study proposes the use of Machine Learning, including Convolutional Neural Networks (CNNs) and classifiers such as Random Forest, Naïve Bayes, Multilayer Perceptron, and Support Vector Machine, to analyze and classify the effects of rain in FSO systems. The methodology involves generating a database through simulations in OptiSystem, followed by preprocessing the images to retain only relevant eye diagrams. A CNN is then applied as a feature extractor, with its attributes used as input for the classifiers. After classification, it is possible to discern the type of rain associated with each entry in the database. The results highlight effective combinations, such as VGG16 and VGG19 with the Bayes classifier for distances of 500m and 1 km, and Random Forest with InceptionV3 for 1 km, achieving accuracies above 90% and 99%, respectively. This study offers a practical and effective approach to signal quality analysis in FSO systems, emphasizing the importance of Machine Learning, especially CNNs, in this context. These techniques allow for precise analysis adaptable to weather conditions, providing valuable insights for future enhancements and real-world implementations.

Systems developed to work with computational intelligence have become very efficient, and in some cases obtain more accurate results than evaluations by humans. Hence, this work proposes a new online approach based on deep learning tools according to the concept of transfer learning to generate a computational intelligence framework for use with the Internet of Health Things (IoHT) devices. This framework allows the user to add their images and perform platform training almost as easily as creating folders and placing files in regular cloud storage services. The trials carried out with the tool showed that even people with no programming and image processing knowledge were able to set up projects in a few minutes. The proposed approach is validated using three medical databases, which include cerebral vascular accident images for stroke type classification, lung nodule images for malignant classification, and skin images for the classification of melanocytic lesions. The results show the efficiency and reliability of the framework, which reached 91.6% Accuracy in the stroke images and lung nodules databases, and 92% Accuracy in the skin images databases. This prove the immense contribution that this work can bring to assist medical professionals in analyzing complex examinations quickly and accurately, allowing a large medical examination database through a consolidated collaborative IoT platform.

The identification of different textile fabrics is a task commonly learned in practice and, therefore, is considered a very strenuous and costly form of learning, causing annoyance to the individual who performs it. Based on this context, this paper proposes a new method for classifying textile fabrics, based on the development of a computer vision system using Convolutional Neural Network (CNN). CNN works as a feature extractor by incorporating the concept of Transfer Learning. Using Transfer Learning allows a pre-trained CNN model to be reused for a new problem. In order to highlight the high performance of CNN, an analysis is performed with feature extractors established in the literature. Parameters such as Accuracy, F1-Score, and processing time are considered to evaluate the efficiency of the proposed approach. For the classification were used Bayesian Classifier, Multi-layer Perceptron (MLP), k-Nearest Neighbor (kNN), Random Forest (RF), and Support Vector Machine (SVM). The results show that the best combination is the CNN architecture DenseNet201 with SVM (RBF), obtaining an accuracy of 94% and F1-Score of 94.2%.

Nutrition is a significant determinant of health, the resolution of many nutritional issues, initially requires an anthropometry examination. Body measures provide data for studying the relationship between diet, nutritional status, and health. Manual and automatic methods can perform body measurements. The manual method usually uses an anthropometric tape. However, the automatic process uses the equipment of Dual-energy X-ray absorptiometry (DXA). Our work presents a new approach to calculate body measures using 2D Camera Images, applying Digital Image Processing, Convolution Neural Networks, and Machine Learning techniques. The dataset used contains 38 exams, for each exam, has four digital images and the dimensions of body measurements, performed by a specialist. The methods used in this work for segmentation were Dense Human Pose Estimation - CNN with the Bayesian, K-Nearest Neighbors, Support Vector Machine, Decision Threes, Adaptive Boosting, Random Forest, Multilayer Perceptron and Expectation-Maximization classifiers. The approach with Dense Human Pose Estimation and Expectation-Maximization reached the best results, with mean squared error (MSE) always bellow 4.606 ± 3.412 cm when compared with specialist measures.

Among illnesses, heart diseases are accounted for as one of the most responsible for deaths. Precise and fast diagnoses increase the patient's chances to receive treatment time. A non-invasive and low-cost way to diagnose it is by using Electrocardiogram (ECG). In this paper, we propose a way to diagnosis two types of heart arrhythmia, by using the ECG record as an image. To access the performance of our system, five feature extraction methods well-known in literature are used along with five different classifiers are tested. We were able to identify heart disorders with over 96.00% of accuracy, using a vanilla neural-network, Multilayer Perceptron (MLP), and Local Binary Patterns (LBP) from ECG images. This investigation has shown promising results from a medical point-of-view.

Lung cancer corresponds to 26% of all deaths due to cancer in 2017, accounting more than 1.5 million deaths globally. Considering this challenging situation, several computeraided diagnosis systems have been developed to detect lung cancer at early stages, which increases the patients' survival rate. Motivated by the success of deep learning in natural and medical image classification tasks, the proposed approach aims to explore the performance of deep transfer learning for lung nodules malignancy classification. For this, convolutional neural networks (CNN), such as VGG16, VGG19, MobileNet, Xception, InceptionV3, ResNet50, Inception-ResNet-V2, DenseNet169, DenseNet201, NASNetMobile and NASNetLarge, were used as features extractors to process the Lung Image Database Consortium and Image Database Resource Initiative (LIDC/IDRI). Next, the deep features returned were classified using Naive Bayes, MultiLayer Perceptron (MLP), Support Vector Machine (SVM), Near Neighbors (KNN) and Random Forest (RF) classifiers. Additionally, to compare the classifiers performance with themselves and with other ones in literature, the evaluation metrics Accuracy (ACC), Area Under the Curve (AUC), True Positive Rate (TPR), Precision (PPV), and F1-Score were computed. Finally, the best combination of deep extractor and classifier was CNN-ResNet50 with SVM-RBF, which achieved ACC of 88.41% and AUC of 93.19%. These results are equivalent to related works, even just using a CNN pre-trained on non-medical images. For this reason, deep transfer learning proved to be a relevant strategy to extract representative imaging biomarkers for lung nodule malignancy classification in chest CT images.

In this paper, we propose a new localization and navigation approach for mobile robots using topological maps and classification with reject option applying convolutional neural networks (CNN) for feature extraction in omnidirectional images. The use of CNN as feature extractor is based on the concept of Transfer Learning. Reject option is used to improve the task of the classifiers, querying information from the topological map. With the objective of evidencing the high performance of the technique considered, an analysis is made between several feature extractors and classifiers, established in the literature. Parameters such as processing time and accuracy are calculated to prove the credibility and effectiveness of the approach, since these properties are fundamental in the analysis of embedded systems. Considering the proposed approach, CNN stands out among the other feature extractors, as it generated the best results in extraction time and accuracy. It obtained an average accuracy of 99.86% and an extraction time of 0.1517s, proving to be a relevant method for the localization and navigation activities.